Film-Tech Forum ARCHIVE: Xpand 3D image problem

This is topic Xpand 3D image problem in forum Digital Cinema Forum at Film-Tech Forum ARCHIVE.

To visit this topic, use this URL:
https://ft-forum.com/ft/cgi-bin/ubb/ultimatebb.cgi?ubb=get_topic;f=16;t=003051

Posted by Gareth Negus (Member # 8729) on 06-22-2017, 01:26 PM:

Our 3D system is Xpand, which we run with an NEC NC800C projector. I've been increasingly noticing a tendency for dark areas of the image to show in indistinct rainbow bands, making chunks of the image blobby and very unpleasant to look at. It doesn't happen every time and sometimes stops if you restart the film.

Has anyone else had this? Is there a likely cause/solution?

Posted by Andrew Carr (Member # 9264) on 06-23-2017, 10:26 AM:

Does it look similar to 8-bit color banding?
Do you notice the same issue if you disable the polarizer and view in 2D? Is it happening on all films?

Posted by Pete Naples (Member # 761) on 06-23-2017, 11:09 AM:

I've seen that with Xpand on a Christie CP2220.

I never did get to the bottom of it, the site retired 3D so it somewhat went away.

Posted by Gareth Negus (Member # 8729) on 06-23-2017, 11:41 AM:

Hi Andrew. Just tried watching the start of Wonder Woman with the 3D switch off and I didn't get the same effect.

It doesn't seem to affect all films (Fantastic Beasts was trouble free) but we show very little 3D so it's hard to say with any certainty if there's a pattern. I'm seeing the same thing on the War for the Planet of the Apes trailer, though The Last Jedi seems fine.

Pete - I'm close to retiring 3D myself (possibly with an axe). It doesn't seem to sell any extra tickets for us and it doesn't seem worth the hassle.

Posted by Pietro Clarici (Member # 4937) on 06-24-2017, 11:03 AM:

I had this a few years ago on an NEC NC2500S.

If I remember correctly, it got noticeably better after increasing the dark time slightly in DCC, and maybe also switching from triple flash - which early Series 1 machines such as ours cannot do at full resolution anyway - to double flash. If you have the credentials to tinker with 3D settings, I'd carefully give that a try.

I also noticed that different types of XpanD glasses behaved differently when it came to banding/posterization: "classic" X101 were the worst, whereas I felt that X103-CP3 and Universal glasses (the ones that you could find in stores and also worked with 3DTVs) did a little better. It probably has something to do with the pi-cell switching speed.

Posted by Peter Castle (Member # 1997) on 06-24-2017, 08:10 PM:

Isn't this just related to color depth.
In 2D, the projector is producing 4:4:2 color, while in 3D this is reduced to 4:2:2?

Posted by Carsten Kurz (Member # 5396) on 06-25-2017, 06:53 AM:

It's only 4:2:2 on legacy (HD-SDI) systems due to signal interface bandwidth constraints. IMB/IMS systems play 2D and 3D in 4:4:4.

I remember there was a complaint like this (banding/rainbow) in another forum years ago. I think they solved it back then as well by adjusting dark time/delay/flashing.

- Carsten

Posted by Pietro Clarici (Member # 4937) on 06-25-2017, 10:49 AM:

Color depth is indeed reduced, but that's not the issue.

It can be demonstrated by configuring the projector in single-channel SDI mode, back in the days we had to do that a few times to show 3D DCPs in 2D. It was 4:2:2, but there was no significant banding. I frankly couldn't tell the difference from a 4:4:4 projection most of the time.

Posted by Marcel Birgelen (Member # 6801) on 06-26-2017, 06:51 AM:

Even with 4:2:2 you shouldn't be able to see banding on screen.

Most banding artifacts occur due to wonky down-conversions, so I think the cause is the down-conversion from 4:4:4 to 4:2:2, happening in the server. It might also be something other going awry, like the "ghostbusting" algorithm, although it should not be operational for shutterglass-based 3D.

Are you doing something else to the image, like are you scaling it?

And are you running a recent firmware image on your server?

Posted by Gareth Negus (Member # 8729) on 06-27-2017, 06:18 AM:

This has very quickly gone above my level of technical expertise! But good to know it's not just me. And I can take notes for the engineer... if he thinks I understand all this he'll have to take it seriously, so thanks all.

Posted by Carsten Kurz (Member # 5396) on 06-27-2017, 07:53 AM:

If it's something in the signal, you would also see it with no 3D glasses on, e.g. when you play 2D content in a 3D playlist. The signal is still 4:2:2 then, or the same 'bad' grading would be visible. But again, you won't see banding due to 4:2:2

The shutter glasses or Z-Screen type modulators cut of the light periodically. As such, they may interfere with the puls-code type light modulation and thus tone modulation of the DLPs. So, I would look for the switch timing of the glasses, delays, dark times, flash rates, etc.

- Carsten

Posted by Marcel Birgelen (Member # 6801) on 06-28-2017, 01:49 AM:

I agree, that if it's in the signal itself, you should still be able to spot it, if you take your 3D glasses off. Obviously, it will be a bit harder to spot, because you now see both images overlayed, but it should still be there.

I think what you're implying is that there is possibly some kind of interference, due to the image being cut-off while it is essentially still being "sequenced" and as such, you might actually lose color depth.

That's quite an interesting theory, but as far as I remember, all TI DLP implementations sequence their colors "continuously", so an early cut-off should primarily result in a darker image, but not directly lead to banding artifacts. Although, if you look at the numbers, you see that you might be operating on a fringe.

TI is pretty sketchy on the exact mirror switching frequency of their chips, but if I remember correctly, it's about 10 kHz for their older generation chips and about 32 kHz for their more recent revisions.

So, let's presume the maximum mirror switching frequency is 10 kHz, at 144 flashes per second, even without dark time, this would mean that there are only about 70 pulses per pixel per image per color. That's even far below the color depth a SVGA graphics cards from the mid to late 1990's could produce in 24 bit "True color".

So, I suspect that if you're tripple-flashing, your problem might be right there. The light engine itself might not have sufficient amount of "pulses per image per pixel" left, to construct an image with sufficient color depth and as such you might see an image with visible banding/posterization artifacts.